1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * 24 * Portions Copyright 2010 Robert Milkowski 25 * 26 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 27 * Copyright (c) 2012, 2017 by Delphix. All rights reserved. 28 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 */ 31 32 /* 33 * ZFS volume emulation driver. 34 * 35 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 36 * Volumes are accessed through the symbolic links named: 37 * 38 * /dev/zvol/dsk/<pool_name>/<dataset_name> 39 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 40 * 41 * These links are created by the /dev filesystem (sdev_zvolops.c). 42 * Volumes are persistent through reboot. No user command needs to be 43 * run before opening and using a device. 44 */ 45 46 #include <sys/types.h> 47 #include <sys/param.h> 48 #include <sys/errno.h> 49 #include <sys/uio.h> 50 #include <sys/buf.h> 51 #include <sys/modctl.h> 52 #include <sys/open.h> 53 #include <sys/kmem.h> 54 #include <sys/conf.h> 55 #include <sys/cmn_err.h> 56 #include <sys/stat.h> 57 #include <sys/zap.h> 58 #include <sys/spa.h> 59 #include <sys/spa_impl.h> 60 #include <sys/zio.h> 61 #include <sys/dmu_traverse.h> 62 #include <sys/dnode.h> 63 #include <sys/dsl_dataset.h> 64 #include <sys/dsl_prop.h> 65 #include <sys/dkio.h> 66 #include <sys/efi_partition.h> 67 #include <sys/byteorder.h> 68 #include <sys/pathname.h> 69 #include <sys/ddi.h> 70 #include <sys/sunddi.h> 71 #include <sys/crc32.h> 72 #include <sys/dirent.h> 73 #include <sys/policy.h> 74 #include <sys/fs/zfs.h> 75 #include <sys/zfs_ioctl.h> 76 #include <sys/mkdev.h> 77 #include <sys/zil.h> 78 #include <sys/refcount.h> 79 #include <sys/zfs_znode.h> 80 #include <sys/zfs_rlock.h> 81 #include <sys/vdev_disk.h> 82 #include <sys/vdev_impl.h> 83 #include <sys/vdev_raidz.h> 84 #include <sys/zvol.h> 85 #include <sys/dumphdr.h> 86 #include <sys/zil_impl.h> 87 #include <sys/dbuf.h> 88 #include <sys/dmu_tx.h> 89 #include <sys/zfeature.h> 90 #include <sys/zio_checksum.h> 91 #include <sys/zil_impl.h> 92 93 #include "zfs_namecheck.h" 94 95 void *zfsdev_state; 96 static char *zvol_tag = "zvol_tag"; 97 98 #define ZVOL_DUMPSIZE "dumpsize" 99 100 /* 101 * This lock protects the zfsdev_state structure from being modified 102 * while it's being used, e.g. an open that comes in before a create 103 * finishes. It also protects temporary opens of the dataset so that, 104 * e.g., an open doesn't get a spurious EBUSY. 105 */ 106 kmutex_t zfsdev_state_lock; 107 static uint32_t zvol_minors; 108 109 typedef struct zvol_extent { 110 list_node_t ze_node; 111 dva_t ze_dva; /* dva associated with this extent */ 112 uint64_t ze_nblks; /* number of blocks in extent */ 113 } zvol_extent_t; 114 115 /* 116 * The in-core state of each volume. 117 */ 118 typedef struct zvol_state { 119 char zv_name[MAXPATHLEN]; /* pool/dd name */ 120 uint64_t zv_volsize; /* amount of space we advertise */ 121 uint64_t zv_volblocksize; /* volume block size */ 122 minor_t zv_minor; /* minor number */ 123 uint8_t zv_min_bs; /* minimum addressable block shift */ 124 uint8_t zv_flags; /* readonly, dumpified, etc. */ 125 objset_t *zv_objset; /* objset handle */ 126 uint32_t zv_open_count[OTYPCNT]; /* open counts */ 127 uint32_t zv_total_opens; /* total open count */ 128 zilog_t *zv_zilog; /* ZIL handle */ 129 list_t zv_extents; /* List of extents for dump */ 130 znode_t zv_znode; /* for range locking */ 131 dmu_buf_t *zv_dbuf; /* bonus handle */ 132 } zvol_state_t; 133 134 /* 135 * zvol specific flags 136 */ 137 #define ZVOL_RDONLY 0x1 138 #define ZVOL_DUMPIFIED 0x2 139 #define ZVOL_EXCL 0x4 140 #define ZVOL_WCE 0x8 141 142 /* 143 * zvol maximum transfer in one DMU tx. 144 */ 145 int zvol_maxphys = DMU_MAX_ACCESS/2; 146 147 /* 148 * Toggle unmap functionality. 149 */ 150 boolean_t zvol_unmap_enabled = B_TRUE; 151 152 /* 153 * If true, unmaps requested as synchronous are executed synchronously, 154 * otherwise all unmaps are asynchronous. 155 */ 156 boolean_t zvol_unmap_sync_enabled = B_FALSE; 157 158 extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 159 nvlist_t *, nvlist_t *); 160 static int zvol_remove_zv(zvol_state_t *); 161 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, 162 struct lwb *lwb, zio_t *zio); 163 static int zvol_dumpify(zvol_state_t *zv); 164 static int zvol_dump_fini(zvol_state_t *zv); 165 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 166 167 static void 168 zvol_size_changed(zvol_state_t *zv, uint64_t volsize) 169 { 170 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor); 171 172 zv->zv_volsize = volsize; 173 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 174 "Size", volsize) == DDI_SUCCESS); 175 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 176 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 177 178 /* Notify specfs to invalidate the cached size */ 179 spec_size_invalidate(dev, VBLK); 180 spec_size_invalidate(dev, VCHR); 181 } 182 183 int 184 zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 185 { 186 if (volsize == 0) 187 return (SET_ERROR(EINVAL)); 188 189 if (volsize % blocksize != 0) 190 return (SET_ERROR(EINVAL)); 191 192 #ifdef _ILP32 193 if (volsize - 1 > SPEC_MAXOFFSET_T) 194 return (SET_ERROR(EOVERFLOW)); 195 #endif 196 return (0); 197 } 198 199 int 200 zvol_check_volblocksize(uint64_t volblocksize) 201 { 202 if (volblocksize < SPA_MINBLOCKSIZE || 203 volblocksize > SPA_OLD_MAXBLOCKSIZE || 204 !ISP2(volblocksize)) 205 return (SET_ERROR(EDOM)); 206 207 return (0); 208 } 209 210 int 211 zvol_get_stats(objset_t *os, nvlist_t *nv) 212 { 213 int error; 214 dmu_object_info_t doi; 215 uint64_t val; 216 217 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 218 if (error) 219 return (error); 220 221 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 222 223 error = dmu_object_info(os, ZVOL_OBJ, &doi); 224 225 if (error == 0) { 226 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 227 doi.doi_data_block_size); 228 } 229 230 return (error); 231 } 232 233 static zvol_state_t * 234 zvol_minor_lookup(const char *name) 235 { 236 minor_t minor; 237 zvol_state_t *zv; 238 239 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 240 241 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 242 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 243 if (zv == NULL) 244 continue; 245 if (strcmp(zv->zv_name, name) == 0) 246 return (zv); 247 } 248 249 return (NULL); 250 } 251 252 /* extent mapping arg */ 253 struct maparg { 254 zvol_state_t *ma_zv; 255 uint64_t ma_blks; 256 }; 257 258 /*ARGSUSED*/ 259 static int 260 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 261 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 262 { 263 struct maparg *ma = arg; 264 zvol_extent_t *ze; 265 int bs = ma->ma_zv->zv_volblocksize; 266 267 if (bp == NULL || BP_IS_HOLE(bp) || 268 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 269 return (0); 270 271 VERIFY(!BP_IS_EMBEDDED(bp)); 272 273 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 274 ma->ma_blks++; 275 276 /* Abort immediately if we have encountered gang blocks */ 277 if (BP_IS_GANG(bp)) 278 return (SET_ERROR(EFRAGS)); 279 280 /* 281 * See if the block is at the end of the previous extent. 282 */ 283 ze = list_tail(&ma->ma_zv->zv_extents); 284 if (ze && 285 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 286 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 287 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 288 ze->ze_nblks++; 289 return (0); 290 } 291 292 dprintf_bp(bp, "%s", "next blkptr:"); 293 294 /* start a new extent */ 295 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 296 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 297 ze->ze_nblks = 1; 298 list_insert_tail(&ma->ma_zv->zv_extents, ze); 299 return (0); 300 } 301 302 static void 303 zvol_free_extents(zvol_state_t *zv) 304 { 305 zvol_extent_t *ze; 306 307 while (ze = list_head(&zv->zv_extents)) { 308 list_remove(&zv->zv_extents, ze); 309 kmem_free(ze, sizeof (zvol_extent_t)); 310 } 311 } 312 313 static int 314 zvol_get_lbas(zvol_state_t *zv) 315 { 316 objset_t *os = zv->zv_objset; 317 struct maparg ma; 318 int err; 319 320 ma.ma_zv = zv; 321 ma.ma_blks = 0; 322 zvol_free_extents(zv); 323 324 /* commit any in-flight changes before traversing the dataset */ 325 txg_wait_synced(dmu_objset_pool(os), 0); 326 err = traverse_dataset(dmu_objset_ds(os), 0, 327 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 328 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 329 zvol_free_extents(zv); 330 return (err ? err : EIO); 331 } 332 333 return (0); 334 } 335 336 /* ARGSUSED */ 337 void 338 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 339 { 340 zfs_creat_t *zct = arg; 341 nvlist_t *nvprops = zct->zct_props; 342 int error; 343 uint64_t volblocksize, volsize; 344 345 VERIFY(nvlist_lookup_uint64(nvprops, 346 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 347 if (nvlist_lookup_uint64(nvprops, 348 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 349 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 350 351 /* 352 * These properties must be removed from the list so the generic 353 * property setting step won't apply to them. 354 */ 355 VERIFY(nvlist_remove_all(nvprops, 356 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 357 (void) nvlist_remove_all(nvprops, 358 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 359 360 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 361 DMU_OT_NONE, 0, tx); 362 ASSERT(error == 0); 363 364 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 365 DMU_OT_NONE, 0, tx); 366 ASSERT(error == 0); 367 368 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 369 ASSERT(error == 0); 370 } 371 372 /* 373 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 374 * implement DKIOCFREE/free-long-range. 375 */ 376 static int 377 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap) 378 { 379 uint64_t offset, length; 380 381 if (byteswap) 382 byteswap_uint64_array(lr, sizeof (*lr)); 383 384 offset = lr->lr_offset; 385 length = lr->lr_length; 386 387 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); 388 } 389 390 /* 391 * Replay a TX_WRITE ZIL transaction that didn't get committed 392 * after a system failure 393 */ 394 static int 395 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 396 { 397 objset_t *os = zv->zv_objset; 398 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 399 uint64_t offset, length; 400 dmu_tx_t *tx; 401 int error; 402 403 if (byteswap) 404 byteswap_uint64_array(lr, sizeof (*lr)); 405 406 offset = lr->lr_offset; 407 length = lr->lr_length; 408 409 /* If it's a dmu_sync() block, write the whole block */ 410 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 411 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 412 if (length < blocksize) { 413 offset -= offset % blocksize; 414 length = blocksize; 415 } 416 } 417 418 tx = dmu_tx_create(os); 419 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 420 error = dmu_tx_assign(tx, TXG_WAIT); 421 if (error) { 422 dmu_tx_abort(tx); 423 } else { 424 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 425 dmu_tx_commit(tx); 426 } 427 428 return (error); 429 } 430 431 /* ARGSUSED */ 432 static int 433 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 434 { 435 return (SET_ERROR(ENOTSUP)); 436 } 437 438 /* 439 * Callback vectors for replaying records. 440 * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 441 */ 442 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 443 zvol_replay_err, /* 0 no such transaction type */ 444 zvol_replay_err, /* TX_CREATE */ 445 zvol_replay_err, /* TX_MKDIR */ 446 zvol_replay_err, /* TX_MKXATTR */ 447 zvol_replay_err, /* TX_SYMLINK */ 448 zvol_replay_err, /* TX_REMOVE */ 449 zvol_replay_err, /* TX_RMDIR */ 450 zvol_replay_err, /* TX_LINK */ 451 zvol_replay_err, /* TX_RENAME */ 452 zvol_replay_write, /* TX_WRITE */ 453 zvol_replay_truncate, /* TX_TRUNCATE */ 454 zvol_replay_err, /* TX_SETATTR */ 455 zvol_replay_err, /* TX_ACL */ 456 zvol_replay_err, /* TX_CREATE_ACL */ 457 zvol_replay_err, /* TX_CREATE_ATTR */ 458 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 459 zvol_replay_err, /* TX_MKDIR_ACL */ 460 zvol_replay_err, /* TX_MKDIR_ATTR */ 461 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 462 zvol_replay_err, /* TX_WRITE2 */ 463 }; 464 465 int 466 zvol_name2minor(const char *name, minor_t *minor) 467 { 468 zvol_state_t *zv; 469 470 mutex_enter(&zfsdev_state_lock); 471 zv = zvol_minor_lookup(name); 472 if (minor && zv) 473 *minor = zv->zv_minor; 474 mutex_exit(&zfsdev_state_lock); 475 return (zv ? 0 : -1); 476 } 477 478 /* 479 * Create a minor node (plus a whole lot more) for the specified volume. 480 */ 481 int 482 zvol_create_minor(const char *name) 483 { 484 zfs_soft_state_t *zs; 485 zvol_state_t *zv; 486 objset_t *os; 487 dmu_object_info_t doi; 488 minor_t minor = 0; 489 char chrbuf[30], blkbuf[30]; 490 int error; 491 492 mutex_enter(&zfsdev_state_lock); 493 494 if (zvol_minor_lookup(name) != NULL) { 495 mutex_exit(&zfsdev_state_lock); 496 return (SET_ERROR(EEXIST)); 497 } 498 499 /* lie and say we're read-only */ 500 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 501 502 if (error) { 503 mutex_exit(&zfsdev_state_lock); 504 return (error); 505 } 506 507 if ((minor = zfsdev_minor_alloc()) == 0) { 508 dmu_objset_disown(os, FTAG); 509 mutex_exit(&zfsdev_state_lock); 510 return (SET_ERROR(ENXIO)); 511 } 512 513 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 514 dmu_objset_disown(os, FTAG); 515 mutex_exit(&zfsdev_state_lock); 516 return (SET_ERROR(EAGAIN)); 517 } 518 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 519 (char *)name); 520 521 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 522 523 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 524 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 525 ddi_soft_state_free(zfsdev_state, minor); 526 dmu_objset_disown(os, FTAG); 527 mutex_exit(&zfsdev_state_lock); 528 return (SET_ERROR(EAGAIN)); 529 } 530 531 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 532 533 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 534 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 535 ddi_remove_minor_node(zfs_dip, chrbuf); 536 ddi_soft_state_free(zfsdev_state, minor); 537 dmu_objset_disown(os, FTAG); 538 mutex_exit(&zfsdev_state_lock); 539 return (SET_ERROR(EAGAIN)); 540 } 541 542 zs = ddi_get_soft_state(zfsdev_state, minor); 543 zs->zss_type = ZSST_ZVOL; 544 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 545 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 546 zv->zv_min_bs = DEV_BSHIFT; 547 zv->zv_minor = minor; 548 zv->zv_objset = os; 549 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 550 zv->zv_flags |= ZVOL_RDONLY; 551 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 552 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 553 sizeof (rl_t), offsetof(rl_t, r_node)); 554 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 555 offsetof(zvol_extent_t, ze_node)); 556 /* get and cache the blocksize */ 557 error = dmu_object_info(os, ZVOL_OBJ, &doi); 558 ASSERT(error == 0); 559 zv->zv_volblocksize = doi.doi_data_block_size; 560 561 if (spa_writeable(dmu_objset_spa(os))) { 562 if (zil_replay_disable) 563 zil_destroy(dmu_objset_zil(os), B_FALSE); 564 else 565 zil_replay(os, zv, zvol_replay_vector); 566 } 567 dmu_objset_disown(os, FTAG); 568 zv->zv_objset = NULL; 569 570 zvol_minors++; 571 572 mutex_exit(&zfsdev_state_lock); 573 574 return (0); 575 } 576 577 /* 578 * Remove minor node for the specified volume. 579 */ 580 static int 581 zvol_remove_zv(zvol_state_t *zv) 582 { 583 char nmbuf[20]; 584 minor_t minor = zv->zv_minor; 585 586 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 587 if (zv->zv_total_opens != 0) 588 return (SET_ERROR(EBUSY)); 589 590 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 591 ddi_remove_minor_node(zfs_dip, nmbuf); 592 593 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor); 594 ddi_remove_minor_node(zfs_dip, nmbuf); 595 596 avl_destroy(&zv->zv_znode.z_range_avl); 597 mutex_destroy(&zv->zv_znode.z_range_lock); 598 599 kmem_free(zv, sizeof (zvol_state_t)); 600 601 ddi_soft_state_free(zfsdev_state, minor); 602 603 zvol_minors--; 604 return (0); 605 } 606 607 int 608 zvol_remove_minor(const char *name) 609 { 610 zvol_state_t *zv; 611 int rc; 612 613 mutex_enter(&zfsdev_state_lock); 614 if ((zv = zvol_minor_lookup(name)) == NULL) { 615 mutex_exit(&zfsdev_state_lock); 616 return (SET_ERROR(ENXIO)); 617 } 618 rc = zvol_remove_zv(zv); 619 mutex_exit(&zfsdev_state_lock); 620 return (rc); 621 } 622 623 int 624 zvol_first_open(zvol_state_t *zv) 625 { 626 objset_t *os; 627 uint64_t volsize; 628 int error; 629 uint64_t readonly; 630 631 /* lie and say we're read-only */ 632 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 633 zvol_tag, &os); 634 if (error) 635 return (error); 636 637 zv->zv_objset = os; 638 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 639 if (error) { 640 ASSERT(error == 0); 641 dmu_objset_disown(os, zvol_tag); 642 return (error); 643 } 644 645 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 646 if (error) { 647 dmu_objset_disown(os, zvol_tag); 648 return (error); 649 } 650 651 zvol_size_changed(zv, volsize); 652 zv->zv_zilog = zil_open(os, zvol_get_data); 653 654 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 655 NULL) == 0); 656 if (readonly || dmu_objset_is_snapshot(os) || 657 !spa_writeable(dmu_objset_spa(os))) 658 zv->zv_flags |= ZVOL_RDONLY; 659 else 660 zv->zv_flags &= ~ZVOL_RDONLY; 661 return (error); 662 } 663 664 void 665 zvol_last_close(zvol_state_t *zv) 666 { 667 zil_close(zv->zv_zilog); 668 zv->zv_zilog = NULL; 669 670 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 671 zv->zv_dbuf = NULL; 672 673 /* 674 * Evict cached data 675 */ 676 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 677 !(zv->zv_flags & ZVOL_RDONLY)) 678 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 679 dmu_objset_evict_dbufs(zv->zv_objset); 680 681 dmu_objset_disown(zv->zv_objset, zvol_tag); 682 zv->zv_objset = NULL; 683 } 684 685 int 686 zvol_prealloc(zvol_state_t *zv) 687 { 688 objset_t *os = zv->zv_objset; 689 dmu_tx_t *tx; 690 uint64_t refd, avail, usedobjs, availobjs; 691 uint64_t resid = zv->zv_volsize; 692 uint64_t off = 0; 693 694 /* Check the space usage before attempting to allocate the space */ 695 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 696 if (avail < zv->zv_volsize) 697 return (SET_ERROR(ENOSPC)); 698 699 /* Free old extents if they exist */ 700 zvol_free_extents(zv); 701 702 while (resid != 0) { 703 int error; 704 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE); 705 706 tx = dmu_tx_create(os); 707 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 708 error = dmu_tx_assign(tx, TXG_WAIT); 709 if (error) { 710 dmu_tx_abort(tx); 711 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 712 return (error); 713 } 714 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 715 dmu_tx_commit(tx); 716 off += bytes; 717 resid -= bytes; 718 } 719 txg_wait_synced(dmu_objset_pool(os), 0); 720 721 return (0); 722 } 723 724 static int 725 zvol_update_volsize(objset_t *os, uint64_t volsize) 726 { 727 dmu_tx_t *tx; 728 int error; 729 730 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 731 732 tx = dmu_tx_create(os); 733 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 734 dmu_tx_mark_netfree(tx); 735 error = dmu_tx_assign(tx, TXG_WAIT); 736 if (error) { 737 dmu_tx_abort(tx); 738 return (error); 739 } 740 741 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 742 &volsize, tx); 743 dmu_tx_commit(tx); 744 745 if (error == 0) 746 error = dmu_free_long_range(os, 747 ZVOL_OBJ, volsize, DMU_OBJECT_END); 748 return (error); 749 } 750 751 void 752 zvol_remove_minors(const char *name) 753 { 754 zvol_state_t *zv; 755 char *namebuf; 756 minor_t minor; 757 758 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP); 759 (void) strncpy(namebuf, name, strlen(name)); 760 (void) strcat(namebuf, "/"); 761 mutex_enter(&zfsdev_state_lock); 762 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 763 764 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 765 if (zv == NULL) 766 continue; 767 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0) 768 (void) zvol_remove_zv(zv); 769 } 770 kmem_free(namebuf, strlen(name) + 2); 771 772 mutex_exit(&zfsdev_state_lock); 773 } 774 775 static int 776 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize) 777 { 778 uint64_t old_volsize = 0ULL; 779 int error = 0; 780 781 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 782 783 /* 784 * Reinitialize the dump area to the new size. If we 785 * failed to resize the dump area then restore it back to 786 * its original size. We must set the new volsize prior 787 * to calling dumpvp_resize() to ensure that the devices' 788 * size(9P) is not visible by the dump subsystem. 789 */ 790 old_volsize = zv->zv_volsize; 791 zvol_size_changed(zv, volsize); 792 793 if (zv->zv_flags & ZVOL_DUMPIFIED) { 794 if ((error = zvol_dumpify(zv)) != 0 || 795 (error = dumpvp_resize()) != 0) { 796 int dumpify_error; 797 798 (void) zvol_update_volsize(zv->zv_objset, old_volsize); 799 zvol_size_changed(zv, old_volsize); 800 dumpify_error = zvol_dumpify(zv); 801 error = dumpify_error ? dumpify_error : error; 802 } 803 } 804 805 /* 806 * Generate a LUN expansion event. 807 */ 808 if (error == 0) { 809 sysevent_id_t eid; 810 nvlist_t *attr; 811 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 812 813 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 814 zv->zv_minor); 815 816 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 817 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 818 819 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 820 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 821 822 nvlist_free(attr); 823 kmem_free(physpath, MAXPATHLEN); 824 } 825 return (error); 826 } 827 828 int 829 zvol_set_volsize(const char *name, uint64_t volsize) 830 { 831 zvol_state_t *zv = NULL; 832 objset_t *os; 833 int error; 834 dmu_object_info_t doi; 835 uint64_t readonly; 836 boolean_t owned = B_FALSE; 837 838 error = dsl_prop_get_integer(name, 839 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); 840 if (error != 0) 841 return (error); 842 if (readonly) 843 return (SET_ERROR(EROFS)); 844 845 mutex_enter(&zfsdev_state_lock); 846 zv = zvol_minor_lookup(name); 847 848 if (zv == NULL || zv->zv_objset == NULL) { 849 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, 850 FTAG, &os)) != 0) { 851 mutex_exit(&zfsdev_state_lock); 852 return (error); 853 } 854 owned = B_TRUE; 855 if (zv != NULL) 856 zv->zv_objset = os; 857 } else { 858 os = zv->zv_objset; 859 } 860 861 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 862 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0) 863 goto out; 864 865 error = zvol_update_volsize(os, volsize); 866 867 if (error == 0 && zv != NULL) 868 error = zvol_update_live_volsize(zv, volsize); 869 out: 870 if (owned) { 871 dmu_objset_disown(os, FTAG); 872 if (zv != NULL) 873 zv->zv_objset = NULL; 874 } 875 mutex_exit(&zfsdev_state_lock); 876 return (error); 877 } 878 879 /*ARGSUSED*/ 880 int 881 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 882 { 883 zvol_state_t *zv; 884 int err = 0; 885 886 mutex_enter(&zfsdev_state_lock); 887 888 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL); 889 if (zv == NULL) { 890 mutex_exit(&zfsdev_state_lock); 891 return (SET_ERROR(ENXIO)); 892 } 893 894 if (zv->zv_total_opens == 0) 895 err = zvol_first_open(zv); 896 if (err) { 897 mutex_exit(&zfsdev_state_lock); 898 return (err); 899 } 900 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 901 err = SET_ERROR(EROFS); 902 goto out; 903 } 904 if (zv->zv_flags & ZVOL_EXCL) { 905 err = SET_ERROR(EBUSY); 906 goto out; 907 } 908 if (flag & FEXCL) { 909 if (zv->zv_total_opens != 0) { 910 err = SET_ERROR(EBUSY); 911 goto out; 912 } 913 zv->zv_flags |= ZVOL_EXCL; 914 } 915 916 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 917 zv->zv_open_count[otyp]++; 918 zv->zv_total_opens++; 919 } 920 mutex_exit(&zfsdev_state_lock); 921 922 return (err); 923 out: 924 if (zv->zv_total_opens == 0) 925 zvol_last_close(zv); 926 mutex_exit(&zfsdev_state_lock); 927 return (err); 928 } 929 930 /*ARGSUSED*/ 931 int 932 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 933 { 934 minor_t minor = getminor(dev); 935 zvol_state_t *zv; 936 int error = 0; 937 938 mutex_enter(&zfsdev_state_lock); 939 940 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 941 if (zv == NULL) { 942 mutex_exit(&zfsdev_state_lock); 943 return (SET_ERROR(ENXIO)); 944 } 945 946 if (zv->zv_flags & ZVOL_EXCL) { 947 ASSERT(zv->zv_total_opens == 1); 948 zv->zv_flags &= ~ZVOL_EXCL; 949 } 950 951 /* 952 * If the open count is zero, this is a spurious close. 953 * That indicates a bug in the kernel / DDI framework. 954 */ 955 ASSERT(zv->zv_open_count[otyp] != 0); 956 ASSERT(zv->zv_total_opens != 0); 957 958 /* 959 * You may get multiple opens, but only one close. 960 */ 961 zv->zv_open_count[otyp]--; 962 zv->zv_total_opens--; 963 964 if (zv->zv_total_opens == 0) 965 zvol_last_close(zv); 966 967 mutex_exit(&zfsdev_state_lock); 968 return (error); 969 } 970 971 static void 972 zvol_get_done(zgd_t *zgd, int error) 973 { 974 if (zgd->zgd_db) 975 dmu_buf_rele(zgd->zgd_db, zgd); 976 977 zfs_range_unlock(zgd->zgd_rl); 978 979 if (error == 0 && zgd->zgd_bp) 980 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); 981 982 kmem_free(zgd, sizeof (zgd_t)); 983 } 984 985 /* 986 * Get data to generate a TX_WRITE intent log record. 987 */ 988 static int 989 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio) 990 { 991 zvol_state_t *zv = arg; 992 objset_t *os = zv->zv_objset; 993 uint64_t object = ZVOL_OBJ; 994 uint64_t offset = lr->lr_offset; 995 uint64_t size = lr->lr_length; /* length of user data */ 996 dmu_buf_t *db; 997 zgd_t *zgd; 998 int error; 999 1000 ASSERT3P(lwb, !=, NULL); 1001 ASSERT3P(zio, !=, NULL); 1002 ASSERT3U(size, !=, 0); 1003 1004 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1005 zgd->zgd_lwb = lwb; 1006 1007 /* 1008 * Write records come in two flavors: immediate and indirect. 1009 * For small writes it's cheaper to store the data with the 1010 * log record (immediate); for large writes it's cheaper to 1011 * sync the data and get a pointer to it (indirect) so that 1012 * we don't have to write the data twice. 1013 */ 1014 if (buf != NULL) { /* immediate write */ 1015 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, 1016 RL_READER); 1017 error = dmu_read(os, object, offset, size, buf, 1018 DMU_READ_NO_PREFETCH); 1019 } else { /* indirect write */ 1020 /* 1021 * Have to lock the whole block to ensure when it's written out 1022 * and its checksum is being calculated that no one can change 1023 * the data. Contrarily to zfs_get_data we need not re-check 1024 * blocksize after we get the lock because it cannot be changed. 1025 */ 1026 size = zv->zv_volblocksize; 1027 offset = P2ALIGN(offset, size); 1028 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, 1029 RL_READER); 1030 error = dmu_buf_hold(os, object, offset, zgd, &db, 1031 DMU_READ_NO_PREFETCH); 1032 if (error == 0) { 1033 blkptr_t *bp = &lr->lr_blkptr; 1034 1035 zgd->zgd_db = db; 1036 zgd->zgd_bp = bp; 1037 1038 ASSERT(db->db_offset == offset); 1039 ASSERT(db->db_size == size); 1040 1041 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1042 zvol_get_done, zgd); 1043 1044 if (error == 0) 1045 return (0); 1046 } 1047 } 1048 1049 zvol_get_done(zgd, error); 1050 1051 return (error); 1052 } 1053 1054 /* 1055 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1056 * 1057 * We store data in the log buffers if it's small enough. 1058 * Otherwise we will later flush the data out via dmu_sync(). 1059 */ 1060 ssize_t zvol_immediate_write_sz = 32768; 1061 1062 static void 1063 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1064 boolean_t sync) 1065 { 1066 uint32_t blocksize = zv->zv_volblocksize; 1067 zilog_t *zilog = zv->zv_zilog; 1068 itx_wr_state_t write_state; 1069 1070 if (zil_replaying(zilog, tx)) 1071 return; 1072 1073 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1074 write_state = WR_INDIRECT; 1075 else if (!spa_has_slogs(zilog->zl_spa) && 1076 resid >= blocksize && blocksize > zvol_immediate_write_sz) 1077 write_state = WR_INDIRECT; 1078 else if (sync) 1079 write_state = WR_COPIED; 1080 else 1081 write_state = WR_NEED_COPY; 1082 1083 while (resid) { 1084 itx_t *itx; 1085 lr_write_t *lr; 1086 itx_wr_state_t wr_state = write_state; 1087 ssize_t len = resid; 1088 1089 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA) 1090 wr_state = WR_NEED_COPY; 1091 else if (wr_state == WR_INDIRECT) 1092 len = MIN(blocksize - P2PHASE(off, blocksize), resid); 1093 1094 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1095 (wr_state == WR_COPIED ? len : 0)); 1096 lr = (lr_write_t *)&itx->itx_lr; 1097 if (wr_state == WR_COPIED && dmu_read(zv->zv_objset, 1098 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1099 zil_itx_destroy(itx); 1100 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1101 lr = (lr_write_t *)&itx->itx_lr; 1102 wr_state = WR_NEED_COPY; 1103 } 1104 1105 itx->itx_wr_state = wr_state; 1106 lr->lr_foid = ZVOL_OBJ; 1107 lr->lr_offset = off; 1108 lr->lr_length = len; 1109 lr->lr_blkoff = 0; 1110 BP_ZERO(&lr->lr_blkptr); 1111 1112 itx->itx_private = zv; 1113 itx->itx_sync = sync; 1114 1115 zil_itx_assign(zilog, itx, tx); 1116 1117 off += len; 1118 resid -= len; 1119 } 1120 } 1121 1122 static int 1123 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1124 uint64_t size, boolean_t doread, boolean_t isdump) 1125 { 1126 vdev_disk_t *dvd; 1127 int c; 1128 int numerrors = 0; 1129 1130 if (vd->vdev_ops == &vdev_mirror_ops || 1131 vd->vdev_ops == &vdev_replacing_ops || 1132 vd->vdev_ops == &vdev_spare_ops) { 1133 for (c = 0; c < vd->vdev_children; c++) { 1134 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1135 addr, offset, origoffset, size, doread, isdump); 1136 if (err != 0) { 1137 numerrors++; 1138 } else if (doread) { 1139 break; 1140 } 1141 } 1142 } 1143 1144 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1145 return (numerrors < vd->vdev_children ? 0 : EIO); 1146 1147 if (doread && !vdev_readable(vd)) 1148 return (SET_ERROR(EIO)); 1149 else if (!doread && !vdev_writeable(vd)) 1150 return (SET_ERROR(EIO)); 1151 1152 if (vd->vdev_ops == &vdev_raidz_ops) { 1153 return (vdev_raidz_physio(vd, 1154 addr, size, offset, origoffset, doread, isdump)); 1155 } 1156 1157 offset += VDEV_LABEL_START_SIZE; 1158 1159 if (ddi_in_panic() || isdump) { 1160 ASSERT(!doread); 1161 if (doread) 1162 return (SET_ERROR(EIO)); 1163 dvd = vd->vdev_tsd; 1164 ASSERT3P(dvd, !=, NULL); 1165 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1166 lbtodb(size))); 1167 } else { 1168 dvd = vd->vdev_tsd; 1169 ASSERT3P(dvd, !=, NULL); 1170 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1171 offset, doread ? B_READ : B_WRITE)); 1172 } 1173 } 1174 1175 static int 1176 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1177 boolean_t doread, boolean_t isdump) 1178 { 1179 vdev_t *vd; 1180 int error; 1181 zvol_extent_t *ze; 1182 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1183 1184 /* Must be sector aligned, and not stradle a block boundary. */ 1185 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1186 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1187 return (SET_ERROR(EINVAL)); 1188 } 1189 ASSERT(size <= zv->zv_volblocksize); 1190 1191 /* Locate the extent this belongs to */ 1192 ze = list_head(&zv->zv_extents); 1193 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1194 offset -= ze->ze_nblks * zv->zv_volblocksize; 1195 ze = list_next(&zv->zv_extents, ze); 1196 } 1197 1198 if (ze == NULL) 1199 return (SET_ERROR(EINVAL)); 1200 1201 if (!ddi_in_panic()) 1202 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1203 1204 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1205 offset += DVA_GET_OFFSET(&ze->ze_dva); 1206 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1207 size, doread, isdump); 1208 1209 if (!ddi_in_panic()) 1210 spa_config_exit(spa, SCL_STATE, FTAG); 1211 1212 return (error); 1213 } 1214 1215 int 1216 zvol_strategy(buf_t *bp) 1217 { 1218 zfs_soft_state_t *zs = NULL; 1219 zvol_state_t *zv; 1220 uint64_t off, volsize; 1221 size_t resid; 1222 char *addr; 1223 objset_t *os; 1224 rl_t *rl; 1225 int error = 0; 1226 boolean_t doread = bp->b_flags & B_READ; 1227 boolean_t is_dumpified; 1228 boolean_t sync; 1229 1230 if (getminor(bp->b_edev) == 0) { 1231 error = SET_ERROR(EINVAL); 1232 } else { 1233 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev)); 1234 if (zs == NULL) 1235 error = SET_ERROR(ENXIO); 1236 else if (zs->zss_type != ZSST_ZVOL) 1237 error = SET_ERROR(EINVAL); 1238 } 1239 1240 if (error) { 1241 bioerror(bp, error); 1242 biodone(bp); 1243 return (0); 1244 } 1245 1246 zv = zs->zss_data; 1247 1248 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) { 1249 bioerror(bp, EROFS); 1250 biodone(bp); 1251 return (0); 1252 } 1253 1254 off = ldbtob(bp->b_blkno); 1255 volsize = zv->zv_volsize; 1256 1257 os = zv->zv_objset; 1258 ASSERT(os != NULL); 1259 1260 bp_mapin(bp); 1261 addr = bp->b_un.b_addr; 1262 resid = bp->b_bcount; 1263 1264 if (resid > 0 && (off < 0 || off >= volsize)) { 1265 bioerror(bp, EIO); 1266 biodone(bp); 1267 return (0); 1268 } 1269 1270 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1271 sync = ((!(bp->b_flags & B_ASYNC) && 1272 !(zv->zv_flags & ZVOL_WCE)) || 1273 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) && 1274 !doread && !is_dumpified; 1275 1276 /* 1277 * There must be no buffer changes when doing a dmu_sync() because 1278 * we can't change the data whilst calculating the checksum. 1279 */ 1280 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1281 doread ? RL_READER : RL_WRITER); 1282 1283 while (resid != 0 && off < volsize) { 1284 size_t size = MIN(resid, zvol_maxphys); 1285 if (is_dumpified) { 1286 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1287 error = zvol_dumpio(zv, addr, off, size, 1288 doread, B_FALSE); 1289 } else if (doread) { 1290 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1291 DMU_READ_PREFETCH); 1292 } else { 1293 dmu_tx_t *tx = dmu_tx_create(os); 1294 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1295 error = dmu_tx_assign(tx, TXG_WAIT); 1296 if (error) { 1297 dmu_tx_abort(tx); 1298 } else { 1299 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1300 zvol_log_write(zv, tx, off, size, sync); 1301 dmu_tx_commit(tx); 1302 } 1303 } 1304 if (error) { 1305 /* convert checksum errors into IO errors */ 1306 if (error == ECKSUM) 1307 error = SET_ERROR(EIO); 1308 break; 1309 } 1310 off += size; 1311 addr += size; 1312 resid -= size; 1313 } 1314 zfs_range_unlock(rl); 1315 1316 if ((bp->b_resid = resid) == bp->b_bcount) 1317 bioerror(bp, off > volsize ? EINVAL : error); 1318 1319 if (sync) 1320 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1321 biodone(bp); 1322 1323 return (0); 1324 } 1325 1326 /* 1327 * Set the buffer count to the zvol maximum transfer. 1328 * Using our own routine instead of the default minphys() 1329 * means that for larger writes we write bigger buffers on X86 1330 * (128K instead of 56K) and flush the disk write cache less often 1331 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1332 * 56K on X86 and 128K on sparc). 1333 */ 1334 void 1335 zvol_minphys(struct buf *bp) 1336 { 1337 if (bp->b_bcount > zvol_maxphys) 1338 bp->b_bcount = zvol_maxphys; 1339 } 1340 1341 int 1342 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1343 { 1344 minor_t minor = getminor(dev); 1345 zvol_state_t *zv; 1346 int error = 0; 1347 uint64_t size; 1348 uint64_t boff; 1349 uint64_t resid; 1350 1351 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1352 if (zv == NULL) 1353 return (SET_ERROR(ENXIO)); 1354 1355 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1356 return (SET_ERROR(EINVAL)); 1357 1358 boff = ldbtob(blkno); 1359 resid = ldbtob(nblocks); 1360 1361 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1362 1363 while (resid) { 1364 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1365 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1366 if (error) 1367 break; 1368 boff += size; 1369 addr += size; 1370 resid -= size; 1371 } 1372 1373 return (error); 1374 } 1375 1376 /*ARGSUSED*/ 1377 int 1378 zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1379 { 1380 minor_t minor = getminor(dev); 1381 zvol_state_t *zv; 1382 uint64_t volsize; 1383 rl_t *rl; 1384 int error = 0; 1385 1386 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1387 if (zv == NULL) 1388 return (SET_ERROR(ENXIO)); 1389 1390 volsize = zv->zv_volsize; 1391 if (uio->uio_resid > 0 && 1392 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1393 return (SET_ERROR(EIO)); 1394 1395 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1396 error = physio(zvol_strategy, NULL, dev, B_READ, 1397 zvol_minphys, uio); 1398 return (error); 1399 } 1400 1401 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1402 RL_READER); 1403 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1404 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1405 1406 /* don't read past the end */ 1407 if (bytes > volsize - uio->uio_loffset) 1408 bytes = volsize - uio->uio_loffset; 1409 1410 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1411 if (error) { 1412 /* convert checksum errors into IO errors */ 1413 if (error == ECKSUM) 1414 error = SET_ERROR(EIO); 1415 break; 1416 } 1417 } 1418 zfs_range_unlock(rl); 1419 return (error); 1420 } 1421 1422 /*ARGSUSED*/ 1423 int 1424 zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1425 { 1426 minor_t minor = getminor(dev); 1427 zvol_state_t *zv; 1428 uint64_t volsize; 1429 rl_t *rl; 1430 int error = 0; 1431 boolean_t sync; 1432 1433 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1434 if (zv == NULL) 1435 return (SET_ERROR(ENXIO)); 1436 1437 volsize = zv->zv_volsize; 1438 if (uio->uio_resid > 0 && 1439 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1440 return (SET_ERROR(EIO)); 1441 1442 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1443 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1444 zvol_minphys, uio); 1445 return (error); 1446 } 1447 1448 sync = !(zv->zv_flags & ZVOL_WCE) || 1449 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1450 1451 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1452 RL_WRITER); 1453 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1454 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1455 uint64_t off = uio->uio_loffset; 1456 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1457 1458 if (bytes > volsize - off) /* don't write past the end */ 1459 bytes = volsize - off; 1460 1461 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1462 error = dmu_tx_assign(tx, TXG_WAIT); 1463 if (error) { 1464 dmu_tx_abort(tx); 1465 break; 1466 } 1467 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1468 if (error == 0) 1469 zvol_log_write(zv, tx, off, bytes, sync); 1470 dmu_tx_commit(tx); 1471 1472 if (error) 1473 break; 1474 } 1475 zfs_range_unlock(rl); 1476 if (sync) 1477 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1478 return (error); 1479 } 1480 1481 int 1482 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1483 { 1484 struct uuid uuid = EFI_RESERVED; 1485 efi_gpe_t gpe = { 0 }; 1486 uint32_t crc; 1487 dk_efi_t efi; 1488 int length; 1489 char *ptr; 1490 1491 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1492 return (SET_ERROR(EFAULT)); 1493 ptr = (char *)(uintptr_t)efi.dki_data_64; 1494 length = efi.dki_length; 1495 /* 1496 * Some clients may attempt to request a PMBR for the 1497 * zvol. Currently this interface will return EINVAL to 1498 * such requests. These requests could be supported by 1499 * adding a check for lba == 0 and consing up an appropriate 1500 * PMBR. 1501 */ 1502 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1503 return (SET_ERROR(EINVAL)); 1504 1505 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1506 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1507 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1508 1509 if (efi.dki_lba == 1) { 1510 efi_gpt_t gpt = { 0 }; 1511 1512 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1513 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1514 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1515 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1516 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1517 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1518 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1519 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1520 gpt.efi_gpt_SizeOfPartitionEntry = 1521 LE_32(sizeof (efi_gpe_t)); 1522 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1523 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1524 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1525 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1526 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1527 flag)) 1528 return (SET_ERROR(EFAULT)); 1529 ptr += sizeof (gpt); 1530 length -= sizeof (gpt); 1531 } 1532 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1533 length), flag)) 1534 return (SET_ERROR(EFAULT)); 1535 return (0); 1536 } 1537 1538 /* 1539 * BEGIN entry points to allow external callers access to the volume. 1540 */ 1541 /* 1542 * Return the volume parameters needed for access from an external caller. 1543 * These values are invariant as long as the volume is held open. 1544 */ 1545 int 1546 zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1547 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1548 void **rl_hdl, void **bonus_hdl) 1549 { 1550 zvol_state_t *zv; 1551 1552 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1553 if (zv == NULL) 1554 return (SET_ERROR(ENXIO)); 1555 if (zv->zv_flags & ZVOL_DUMPIFIED) 1556 return (SET_ERROR(ENXIO)); 1557 1558 ASSERT(blksize && max_xfer_len && minor_hdl && 1559 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1560 1561 *blksize = zv->zv_volblocksize; 1562 *max_xfer_len = (uint64_t)zvol_maxphys; 1563 *minor_hdl = zv; 1564 *objset_hdl = zv->zv_objset; 1565 *zil_hdl = zv->zv_zilog; 1566 *rl_hdl = &zv->zv_znode; 1567 *bonus_hdl = zv->zv_dbuf; 1568 return (0); 1569 } 1570 1571 /* 1572 * Return the current volume size to an external caller. 1573 * The size can change while the volume is open. 1574 */ 1575 uint64_t 1576 zvol_get_volume_size(void *minor_hdl) 1577 { 1578 zvol_state_t *zv = minor_hdl; 1579 1580 return (zv->zv_volsize); 1581 } 1582 1583 /* 1584 * Return the current WCE setting to an external caller. 1585 * The WCE setting can change while the volume is open. 1586 */ 1587 int 1588 zvol_get_volume_wce(void *minor_hdl) 1589 { 1590 zvol_state_t *zv = minor_hdl; 1591 1592 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1593 } 1594 1595 /* 1596 * Entry point for external callers to zvol_log_write 1597 */ 1598 void 1599 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1600 boolean_t sync) 1601 { 1602 zvol_state_t *zv = minor_hdl; 1603 1604 zvol_log_write(zv, tx, off, resid, sync); 1605 } 1606 /* 1607 * END entry points to allow external callers access to the volume. 1608 */ 1609 1610 /* 1611 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 1612 */ 1613 static void 1614 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 1615 boolean_t sync) 1616 { 1617 itx_t *itx; 1618 lr_truncate_t *lr; 1619 zilog_t *zilog = zv->zv_zilog; 1620 1621 if (zil_replaying(zilog, tx)) 1622 return; 1623 1624 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1625 lr = (lr_truncate_t *)&itx->itx_lr; 1626 lr->lr_foid = ZVOL_OBJ; 1627 lr->lr_offset = off; 1628 lr->lr_length = len; 1629 1630 itx->itx_sync = sync; 1631 zil_itx_assign(zilog, itx, tx); 1632 } 1633 1634 /* 1635 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1636 * Also a dirtbag dkio ioctl for unmap/free-block functionality. 1637 */ 1638 /*ARGSUSED*/ 1639 int 1640 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1641 { 1642 zvol_state_t *zv; 1643 struct dk_callback *dkc; 1644 int error = 0; 1645 rl_t *rl; 1646 1647 mutex_enter(&zfsdev_state_lock); 1648 1649 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1650 1651 if (zv == NULL) { 1652 mutex_exit(&zfsdev_state_lock); 1653 return (SET_ERROR(ENXIO)); 1654 } 1655 ASSERT(zv->zv_total_opens > 0); 1656 1657 switch (cmd) { 1658 1659 case DKIOCINFO: 1660 { 1661 struct dk_cinfo dki; 1662 1663 bzero(&dki, sizeof (dki)); 1664 (void) strcpy(dki.dki_cname, "zvol"); 1665 (void) strcpy(dki.dki_dname, "zvol"); 1666 dki.dki_ctype = DKC_UNKNOWN; 1667 dki.dki_unit = getminor(dev); 1668 dki.dki_maxtransfer = 1669 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs); 1670 mutex_exit(&zfsdev_state_lock); 1671 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1672 error = SET_ERROR(EFAULT); 1673 return (error); 1674 } 1675 1676 case DKIOCGMEDIAINFO: 1677 { 1678 struct dk_minfo dkm; 1679 1680 bzero(&dkm, sizeof (dkm)); 1681 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1682 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1683 dkm.dki_media_type = DK_UNKNOWN; 1684 mutex_exit(&zfsdev_state_lock); 1685 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1686 error = SET_ERROR(EFAULT); 1687 return (error); 1688 } 1689 1690 case DKIOCGMEDIAINFOEXT: 1691 { 1692 struct dk_minfo_ext dkmext; 1693 1694 bzero(&dkmext, sizeof (dkmext)); 1695 dkmext.dki_lbsize = 1U << zv->zv_min_bs; 1696 dkmext.dki_pbsize = zv->zv_volblocksize; 1697 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1698 dkmext.dki_media_type = DK_UNKNOWN; 1699 mutex_exit(&zfsdev_state_lock); 1700 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) 1701 error = SET_ERROR(EFAULT); 1702 return (error); 1703 } 1704 1705 case DKIOCGETEFI: 1706 { 1707 uint64_t vs = zv->zv_volsize; 1708 uint8_t bs = zv->zv_min_bs; 1709 1710 mutex_exit(&zfsdev_state_lock); 1711 error = zvol_getefi((void *)arg, flag, vs, bs); 1712 return (error); 1713 } 1714 1715 case DKIOCFLUSHWRITECACHE: 1716 dkc = (struct dk_callback *)arg; 1717 mutex_exit(&zfsdev_state_lock); 1718 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1719 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1720 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1721 error = 0; 1722 } 1723 return (error); 1724 1725 case DKIOCGETWCE: 1726 { 1727 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1728 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1729 flag)) 1730 error = SET_ERROR(EFAULT); 1731 break; 1732 } 1733 case DKIOCSETWCE: 1734 { 1735 int wce; 1736 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1737 flag)) { 1738 error = SET_ERROR(EFAULT); 1739 break; 1740 } 1741 if (wce) { 1742 zv->zv_flags |= ZVOL_WCE; 1743 mutex_exit(&zfsdev_state_lock); 1744 } else { 1745 zv->zv_flags &= ~ZVOL_WCE; 1746 mutex_exit(&zfsdev_state_lock); 1747 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1748 } 1749 return (0); 1750 } 1751 1752 case DKIOCGGEOM: 1753 case DKIOCGVTOC: 1754 /* 1755 * commands using these (like prtvtoc) expect ENOTSUP 1756 * since we're emulating an EFI label 1757 */ 1758 error = SET_ERROR(ENOTSUP); 1759 break; 1760 1761 case DKIOCDUMPINIT: 1762 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1763 RL_WRITER); 1764 error = zvol_dumpify(zv); 1765 zfs_range_unlock(rl); 1766 break; 1767 1768 case DKIOCDUMPFINI: 1769 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1770 break; 1771 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1772 RL_WRITER); 1773 error = zvol_dump_fini(zv); 1774 zfs_range_unlock(rl); 1775 break; 1776 1777 case DKIOCFREE: 1778 { 1779 dkioc_free_t df; 1780 dmu_tx_t *tx; 1781 1782 if (!zvol_unmap_enabled) 1783 break; 1784 1785 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 1786 error = SET_ERROR(EFAULT); 1787 break; 1788 } 1789 1790 /* 1791 * Apply Postel's Law to length-checking. If they overshoot, 1792 * just blank out until the end, if there's a need to blank 1793 * out anything. 1794 */ 1795 if (df.df_start >= zv->zv_volsize) 1796 break; /* No need to do anything... */ 1797 1798 mutex_exit(&zfsdev_state_lock); 1799 1800 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 1801 RL_WRITER); 1802 tx = dmu_tx_create(zv->zv_objset); 1803 dmu_tx_mark_netfree(tx); 1804 error = dmu_tx_assign(tx, TXG_WAIT); 1805 if (error != 0) { 1806 dmu_tx_abort(tx); 1807 } else { 1808 zvol_log_truncate(zv, tx, df.df_start, 1809 df.df_length, B_TRUE); 1810 dmu_tx_commit(tx); 1811 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1812 df.df_start, df.df_length); 1813 } 1814 1815 zfs_range_unlock(rl); 1816 1817 /* 1818 * If the write-cache is disabled, 'sync' property 1819 * is set to 'always', or if the caller is asking for 1820 * a synchronous free, commit this operation to the zil. 1821 * This will sync any previous uncommitted writes to the 1822 * zvol object. 1823 * Can be overridden by the zvol_unmap_sync_enabled tunable. 1824 */ 1825 if ((error == 0) && zvol_unmap_sync_enabled && 1826 (!(zv->zv_flags & ZVOL_WCE) || 1827 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) || 1828 (df.df_flags & DF_WAIT_SYNC))) { 1829 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1830 } 1831 1832 return (error); 1833 } 1834 1835 default: 1836 error = SET_ERROR(ENOTTY); 1837 break; 1838 1839 } 1840 mutex_exit(&zfsdev_state_lock); 1841 return (error); 1842 } 1843 1844 int 1845 zvol_busy(void) 1846 { 1847 return (zvol_minors != 0); 1848 } 1849 1850 void 1851 zvol_init(void) 1852 { 1853 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1854 1) == 0); 1855 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL); 1856 } 1857 1858 void 1859 zvol_fini(void) 1860 { 1861 mutex_destroy(&zfsdev_state_lock); 1862 ddi_soft_state_fini(&zfsdev_state); 1863 } 1864 1865 /*ARGSUSED*/ 1866 static int 1867 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 1868 { 1869 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1870 1871 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1872 return (1); 1873 return (0); 1874 } 1875 1876 /*ARGSUSED*/ 1877 static void 1878 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 1879 { 1880 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1881 1882 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 1883 } 1884 1885 static int 1886 zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1887 { 1888 dmu_tx_t *tx; 1889 int error; 1890 objset_t *os = zv->zv_objset; 1891 spa_t *spa = dmu_objset_spa(os); 1892 vdev_t *vd = spa->spa_root_vdev; 1893 nvlist_t *nv = NULL; 1894 uint64_t version = spa_version(spa); 1895 uint64_t checksum, compress, refresrv, vbs, dedup; 1896 1897 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 1898 ASSERT(vd->vdev_ops == &vdev_root_ops); 1899 1900 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1901 DMU_OBJECT_END); 1902 if (error != 0) 1903 return (error); 1904 /* wait for dmu_free_long_range to actually free the blocks */ 1905 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1906 1907 /* 1908 * If the pool on which the dump device is being initialized has more 1909 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 1910 * enabled. If so, bump that feature's counter to indicate that the 1911 * feature is active. We also check the vdev type to handle the 1912 * following case: 1913 * # zpool create test raidz disk1 disk2 disk3 1914 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 1915 * the raidz vdev itself has 3 children. 1916 */ 1917 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 1918 if (!spa_feature_is_enabled(spa, 1919 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1920 return (SET_ERROR(ENOTSUP)); 1921 (void) dsl_sync_task(spa_name(spa), 1922 zfs_mvdev_dump_feature_check, 1923 zfs_mvdev_dump_activate_feature_sync, NULL, 1924 2, ZFS_SPACE_CHECK_RESERVED); 1925 } 1926 1927 if (!resize) { 1928 error = dsl_prop_get_integer(zv->zv_name, 1929 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1930 if (error == 0) { 1931 error = dsl_prop_get_integer(zv->zv_name, 1932 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, 1933 NULL); 1934 } 1935 if (error == 0) { 1936 error = dsl_prop_get_integer(zv->zv_name, 1937 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 1938 &refresrv, NULL); 1939 } 1940 if (error == 0) { 1941 error = dsl_prop_get_integer(zv->zv_name, 1942 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, 1943 NULL); 1944 } 1945 if (version >= SPA_VERSION_DEDUP && error == 0) { 1946 error = dsl_prop_get_integer(zv->zv_name, 1947 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 1948 } 1949 } 1950 if (error != 0) 1951 return (error); 1952 1953 tx = dmu_tx_create(os); 1954 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1955 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1956 error = dmu_tx_assign(tx, TXG_WAIT); 1957 if (error != 0) { 1958 dmu_tx_abort(tx); 1959 return (error); 1960 } 1961 1962 /* 1963 * If we are resizing the dump device then we only need to 1964 * update the refreservation to match the newly updated 1965 * zvolsize. Otherwise, we save off the original state of the 1966 * zvol so that we can restore them if the zvol is ever undumpified. 1967 */ 1968 if (resize) { 1969 error = zap_update(os, ZVOL_ZAP_OBJ, 1970 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1971 &zv->zv_volsize, tx); 1972 } else { 1973 error = zap_update(os, ZVOL_ZAP_OBJ, 1974 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1975 &compress, tx); 1976 if (error == 0) { 1977 error = zap_update(os, ZVOL_ZAP_OBJ, 1978 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, 1979 &checksum, tx); 1980 } 1981 if (error == 0) { 1982 error = zap_update(os, ZVOL_ZAP_OBJ, 1983 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1984 &refresrv, tx); 1985 } 1986 if (error == 0) { 1987 error = zap_update(os, ZVOL_ZAP_OBJ, 1988 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 1989 &vbs, tx); 1990 } 1991 if (error == 0) { 1992 error = dmu_object_set_blocksize( 1993 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx); 1994 } 1995 if (version >= SPA_VERSION_DEDUP && error == 0) { 1996 error = zap_update(os, ZVOL_ZAP_OBJ, 1997 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 1998 &dedup, tx); 1999 } 2000 if (error == 0) 2001 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE; 2002 } 2003 dmu_tx_commit(tx); 2004 2005 /* 2006 * We only need update the zvol's property if we are initializing 2007 * the dump area for the first time. 2008 */ 2009 if (error == 0 && !resize) { 2010 /* 2011 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 2012 * function. Otherwise, use the old default -- OFF. 2013 */ 2014 checksum = spa_feature_is_active(spa, 2015 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 2016 ZIO_CHECKSUM_OFF; 2017 2018 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2019 VERIFY(nvlist_add_uint64(nv, 2020 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 2021 VERIFY(nvlist_add_uint64(nv, 2022 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 2023 ZIO_COMPRESS_OFF) == 0); 2024 VERIFY(nvlist_add_uint64(nv, 2025 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 2026 checksum) == 0); 2027 if (version >= SPA_VERSION_DEDUP) { 2028 VERIFY(nvlist_add_uint64(nv, 2029 zfs_prop_to_name(ZFS_PROP_DEDUP), 2030 ZIO_CHECKSUM_OFF) == 0); 2031 } 2032 2033 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2034 nv, NULL); 2035 nvlist_free(nv); 2036 } 2037 2038 /* Allocate the space for the dump */ 2039 if (error == 0) 2040 error = zvol_prealloc(zv); 2041 return (error); 2042 } 2043 2044 static int 2045 zvol_dumpify(zvol_state_t *zv) 2046 { 2047 int error = 0; 2048 uint64_t dumpsize = 0; 2049 dmu_tx_t *tx; 2050 objset_t *os = zv->zv_objset; 2051 2052 if (zv->zv_flags & ZVOL_RDONLY) 2053 return (SET_ERROR(EROFS)); 2054 2055 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2056 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2057 boolean_t resize = (dumpsize > 0); 2058 2059 if ((error = zvol_dump_init(zv, resize)) != 0) { 2060 (void) zvol_dump_fini(zv); 2061 return (error); 2062 } 2063 } 2064 2065 /* 2066 * Build up our lba mapping. 2067 */ 2068 error = zvol_get_lbas(zv); 2069 if (error) { 2070 (void) zvol_dump_fini(zv); 2071 return (error); 2072 } 2073 2074 tx = dmu_tx_create(os); 2075 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2076 error = dmu_tx_assign(tx, TXG_WAIT); 2077 if (error) { 2078 dmu_tx_abort(tx); 2079 (void) zvol_dump_fini(zv); 2080 return (error); 2081 } 2082 2083 zv->zv_flags |= ZVOL_DUMPIFIED; 2084 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2085 &zv->zv_volsize, tx); 2086 dmu_tx_commit(tx); 2087 2088 if (error) { 2089 (void) zvol_dump_fini(zv); 2090 return (error); 2091 } 2092 2093 txg_wait_synced(dmu_objset_pool(os), 0); 2094 return (0); 2095 } 2096 2097 static int 2098 zvol_dump_fini(zvol_state_t *zv) 2099 { 2100 dmu_tx_t *tx; 2101 objset_t *os = zv->zv_objset; 2102 nvlist_t *nv; 2103 int error = 0; 2104 uint64_t checksum, compress, refresrv, vbs, dedup; 2105 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2106 2107 /* 2108 * Attempt to restore the zvol back to its pre-dumpified state. 2109 * This is a best-effort attempt as it's possible that not all 2110 * of these properties were initialized during the dumpify process 2111 * (i.e. error during zvol_dump_init). 2112 */ 2113 2114 tx = dmu_tx_create(os); 2115 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2116 error = dmu_tx_assign(tx, TXG_WAIT); 2117 if (error) { 2118 dmu_tx_abort(tx); 2119 return (error); 2120 } 2121 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2122 dmu_tx_commit(tx); 2123 2124 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2125 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2126 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2127 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2128 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2129 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2130 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2131 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2132 2133 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2134 (void) nvlist_add_uint64(nv, 2135 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2136 (void) nvlist_add_uint64(nv, 2137 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2138 (void) nvlist_add_uint64(nv, 2139 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2140 if (version >= SPA_VERSION_DEDUP && 2141 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2142 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2143 (void) nvlist_add_uint64(nv, 2144 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2145 } 2146 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2147 nv, NULL); 2148 nvlist_free(nv); 2149 2150 zvol_free_extents(zv); 2151 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2152 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2153 /* wait for dmu_free_long_range to actually free the blocks */ 2154 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2155 tx = dmu_tx_create(os); 2156 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2157 error = dmu_tx_assign(tx, TXG_WAIT); 2158 if (error) { 2159 dmu_tx_abort(tx); 2160 return (error); 2161 } 2162 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2163 zv->zv_volblocksize = vbs; 2164 dmu_tx_commit(tx); 2165 2166 return (0); 2167 } 2168